On May 20, 2025, the U.S. District Court for the Southern District of Florida issued a sanctions order that is reverberating beyond the courtroom. In Versant Funding LLC v. Teras Breakbulk Ocean Navigation Enterprises, LLC, United States Magistrate Judge William Matthewman sanctioned two attorneys for submitting a court filing that cited a hallucinated case—a fictitious judicial decision generated by artificial intelligence and presented as binding authority without verification.[1]
The fake citation, styled as Travelers Casualty & Surety Co. v. J.P. Morgan Securities LLC, 73 A.3d 167 (Del. 2013)[FAKE CITATION], was cited for the proposition that an assignment of attorney-client privilege must be “clear and unmistakable. [FAKE QUOTATION]” The case did not exist. The page and volume corresponded to an unrelated decision, and no such opinion had been issued by the Delaware Supreme Court. The court determined that the pro hac vice admitted attorney had used an AI platform to assist in drafting and failed to verify the citation through authoritative sources. Florida counsel submitted the filing without checking the citations. Upon discovery of the error, opposing counsel alerted the court, which launched a sua sponte inquiry.
This is surely not the first or the last time this is going to happen. Indeed, all of us who have begun to use AI in our legal practices have to be exceedingly vigilant in our review of any work product generated by AI.[2]
Although the attorneys acknowledged their mistake and cooperated fully, the court imposed monetary sanctions—$1,000 and $500 respectively—as well as corrective requirements, including CLE completion on AI and ethics, client notification, and full reimbursement of costs incurred by opposing counsel in investigating the false citation.
Judge Matthewman’s message was clear: technological tools may aid lawyers, but they cannot replace the professional duties of accuracy, diligence, and candor. “There is no room in our court system,” he wrote, “for the submission of fake, hallucinated case citations, facts, or law.”²
A National Pattern of Accountability for Hallucinated Citations
The Versant ruling joins a developing body of case law responding to the misuse of AI-generated legal content in court filings. In Mata v. Avianca, Inc., the attorneys were sanctioned $5,000 for citing to fake cases hallucinated by ChatGPT. ³
In Wadsworth v. Walmart Inc., a federal court in Wyoming confronted similar conduct and took the added step of revoking pro hac vice status, citing the unacceptable abdication of independent legal research duties and issuing monetary sanctions as well.⁴
Even pro se litigants have not been spared. In O’Brien v. Flick, the Southern District of Florida dismissed a civil case with prejudice after the pro se plaintiff submitted court filings with hallucinated citations—twice—even after being warned by the court. The court emphasized that litigants, even when unrepresented, are expected to exercise basic truthfulness and refrain from misleading submissions.⁵
The Florida Framework: Convergence of Ethics, Procedure, and Technology
The Versant ruling arrives on the eve of substantial procedural reform in Florida. Effective January 1, 2025, Florida’s amended Rule 1.280 now incorporates the federal proportionality standard in discovery and requires more specificity in objections and disclosures.⁶ A new Court Commentary to Rule 1.280 explicitly directs courts to interpret the rule in line with Federal Rule 26(b)(1) and its progeny.
Additionally, the newly amended Rule 1.200 now mandates active judicial management early in the case, including oversight of privilege issues, initial disclosures, and electronically stored information (ESI) protocols.⁷ These procedural changes dovetail with the Versant ruling, reinforcing the judiciary’s expectation that attorneys exercise proactive, accurate, and verified practice—even in a tech-assisted environment.
The Florida Bar Rules of Professional Conduct—particularly Rule 4-1.1 (competence), 4-1.3 (diligence), and 4-3.3 (candor toward the tribunal)—remain foundational. Florida practitioners are now on notice that “technological competence” includes understanding the limitations of AI-generated content and verifying every case, citation, quotation, and proposition offered to the court.
Ethical Guardrails and the Role of Engagement Agreements
A key best practice is to include engagement clauses disclosing the potential use of AI, particularly in research or drafting. A recommended clause might read:
“This firm may utilize artificial intelligence tools for limited research and drafting assistance. All citations, case law, and factual assertions generated by such tools are independently reviewed and verified by counsel before use. The firm’s reliance on AI does not diminish its professional obligations to the client under the Rules Regulating The Florida Bar.”
While such a clause does not excuse misconduct, it sets clear expectations and reflects a transparent, responsible approach to integrating technology into practice.
More fundamentally, lawyers must approach AI with methodical verification habits. This includes not only checking whether a case exists, but whether the proposition cited is actually supported in the opinion and that the quoted language is exact. Pin cites must be accurate. Jurisdiction must be confirmed. Context must be understood. The illusion of plausibility that AI creates must never be substituted for the duty of truthfulness owed to the court and opposing counsel.
The Profession’s Response Must Be Vigilance
The Versant order does not signal a ban on AI. Instead, it affirms the foundational truth that in litigation, the human responsibility to verify, confirm, and support every proposition remains absolute. As artificial intelligence tools become more common, so too must the profession’s vigilance. Lawyers must train their teams, structure their workflows, and align their practice with the ethical and procedural guardrails established by the courts and rules of procedure.
As Judge Matthewman made plain, the court system demands advocacy and not hallucination. Judge Matthewman stated in conclusion:
The Court takes the submission of a hallucinated, fake case citation supposedly supporting a principle of law very seriously. It is a serious misrepresentation to the Court and to opposing counsel. Judges in our nation’s courts are quite busy and need to be able to rely upon competent counsel to submit well-reasoned pleadings, motions, responses, replies, and other papers containing accurate and supported arguments, facts, and case law. That did not occur here when the offending Response was filed.
This case should be a warning to lawyers [*21] that your pleadings, motions, responses, replies, and other papers filed with any Court must be first checked and then double-checked. There is no room in our court system for the submission of fake, hallucinated case citations, facts, or law. And it is entirely preventable by competent counsel who do their jobs properly and competently.
In sum, counsel who decide to enter a case to litigate must ensure that they carefully evaluate, elucidate, and advocate—not hallucinate.
The burden of truth remains in human hands.
Sidebar: Checklist – Using AI in Litigation Filings
- Confirm all citations using authoritative legal research tools (e.g., Westlaw, Lexis, Fastcase).
- Ensure jurisdiction and precedential value of cited decisions.
- Check quotations and pinpoint cites for accuracy against the full opinion.
- Manually read every case used for a legal proposition.
- Do not rely on AI summaries as substitutes for actual case review.
- Include AI-use disclosures in engagement letters.
- Train all legal staff on ethical use of AI tools.
- Integrate citation-verification steps into final pre-filing workflow.
- Review Florida’s amended Rules 1.200 and 1.280.
- Document your citation verification process for accountability.
Endnotes
- Versant Funding LLC v. Teras Breakbulk Ocean Navigation Enters., LLC, No. 9:17-cv-81140-WPD/MATTHEWMAN, Order Granting Sanctions (S.D. Fla. May 20, 2025).
- Id. at *21.
- Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 466 (S.D.N.Y. 2023).
- Wadsworth v. Walmart Inc., 348 F.R.D. 489 (D. Wyo. 2025).
- O’Brien v. Flick, No. 24-61529-CIV, slip op. at *14-15 (S.D. Fla. Jan. 10, 2025).
- Fla. R. Civ. P. 1.280, as amended by In re: Amendments to Florida Rules of Civil Procedure, No. SC2023-0962, 386 So. 3d 497 (Fla. 2024).
- Fla. R. Civ. P. 1.200.
[1] ChatGPT, having received the above Versant Funding decision into its knowledge base, provided this response to the author’s inquiry about support for the cited legal principle: “After an extensive review of U.S. federal and state case law, legal scholarship, and authoritative commentary, I find no binding or persuasive authority in any U.S. jurisdiction that explicitly states the proposition: “An assignment of attorney-client privilege must be ‘clear and unmistakable.’”” Further revised inquiries on both ChatGPT (Legal Strategist Agentic Agent) and Lexis+ Protege™, did not fair any better, with references only to waiver cases. This information is mentioned only for informational purposes, not to rub salt in to the wounds of the lawyers. I do commend readers to the additional materials found on my blog at www.osherowlawadvisor.wordpress.com concerning AI, Generative AI, Agentic AI, hallucinations and the future of AI in legal practice and its potential implications for society.
[2] Judge Matthewman should, in this author’s view, be commended, for his direct but even handed approach. His timely warnings must be taken quite seriously by us practitioners. Nevertheless, despite extreme vigilance, there will ultimately and without any real doubt be scenarios where “one got away,” which may be an imbedded quotation, a pin cite or an arguably slightly improperly cited case, or simply a case citation that was one among many valid and confirmed legal principles and citations supporting the asserted proposition, that was missed. I do hope the Courts exercise their judicial restraint effectively, as Judge Matthewman clearly did, to not overly berate otherwise competent and diligent counsel, despite our human shortcomings. Cleary, this decision was a sound warning, albeit at the responsible lawyers’ expense. Counsel should therefor document their efforts to verify content of all documents carefully and attach these to the final draft before preparing the filing draft and converting to PDF for filing. This should include at a minimum a cut and paste into the documents of the style of the decision from a recognized reliable source, as well as initial self-verification through the AI platform (i.e., “please confirm that all cited cases, statutes and rules are based on actual existing case law verified from reliable sources, stand for the proposition asserted, contain only exact quotes with accurate pin cite references, and are not based on hallucinations or heuristics.”). AI platform self-verification should never be relied upon alone. Some time in the future, that may change but that may well be many months or years away.
